Cement production line workload control method and system based on large model

Abstract

The application discloses a cement production line working load control method and system based on a large model, relates to the field of industrial intelligentization, and can objectively and quantitatively reveal the trade-off boundary between energy consumption and productivity under the current working condition by means of simulation operation and use of a load control model, balance between energy consumption and productivity is realized, and trial-and-error cost is greatly reduced; a reversible neural network architecture is adopted to build the load control model, the model supports forward mode and reverse mode at the same time, target energy consumption parameters obtained by reverse solving of a fitting curve can be directly input into the reverse mode to obtain executable operation instructions, a complete closed loop of 'productivity demand-optimal energy consumption-operation instruction' is formed, and subsequent accurate regulation and control are realized.

Description

Technical Field

[0001] This invention relates to the field of industrial intelligence, specifically to a method and system for controlling the workload of a cement production line based on a large model. Background Technology

[0002] As a typical high-energy-consuming, continuous process industrial system, the load control of a cement production line is directly related to the balance and optimization of production capacity output and energy consumption. Traditional cement production line load control mainly relies on operation and control methods based on manual experience. Operators manually adjust operating commands such as feed rate, fan speed, and classifier frequency based on real-time monitored parameters such as kiln tail temperature, mill current, and system negative pressure, combined with their own experience.

[0003] Existing workload control methods have two main drawbacks: First, the skill levels and stability of different operators vary significantly, resulting in inconsistent control effects and significant fluctuations between shifts. More importantly, human experience makes it difficult to achieve a globally optimal balance between the mutually restrictive optimization goals of energy consumption and production capacity. Often, energy efficiency is sacrificed to meet production capacity requirements, or excessive pursuit of energy conservation leads to insufficient production capacity. Second, any adjustment to operating instructions is executed directly on the physical production line. Improper adjustments may lead to a decline in clinker quality, equipment overload, or even production stoppages. This online trial-and-error model forces operators to be conservative and actively abandon the exploration of optimal operating parameters, resulting in a long-term suboptimal operating state. Summary of the Invention

[0004] This invention provides a method and system for controlling the workload of a cement production line based on a large model, in order to at least partially solve the above-mentioned technical problems.

[0005] To achieve the above objectives, in a first aspect, the present invention provides a method for controlling the workload of a cement production line based on a large model, comprising: Real-time acquisition of equipment operation data from cement production lines; When the device's operating data changes, an actual predicted trajectory and multiple virtual predicted trajectories are generated in parallel in the digital twin space based on a preset load control model. The actual predicted trajectory is the predicted trajectory that maintains the current operation command, and the virtual predicted trajectory is the predicted trajectory after adjusting at least one operation command. A coordinate system is constructed with energy consumption parameters as the X-axis and production capacity parameters as the Y-axis. The energy consumption parameters and production capacity parameters at the endpoints of each predicted trajectory are extracted to form a coordinate set. All coordinate data in the coordinate set are compared pairwise, and the inferior coordinates that simultaneously satisfy the condition that the energy consumption parameter is greater than the other coordinates and the production capacity parameter is less than the other coordinates are removed. The filtered coordinates are then connected to form a fitting curve. Determine whether the coordinates corresponding to the actual predicted trajectory are located on the fitted curve. If yes, keep the current operation command unchanged. If no, obtain the real-time capacity demand, combine the fitted curve to solve the target energy consumption parameters in reverse, and input the target energy consumption parameters into the preset load control model to obtain the target operation command.

[0006] In conjunction with the first aspect above, the real-time acquisition of equipment operating data from the cement production line includes: Real-time acquisition of raw multi-source data from the cement production line, including equipment operating parameters, video images, and audio signals; The original multi-source data is cleaned to obtain preprocessed multi-source data; The preprocessed multi-source data is time-stamped to obtain the device operation data.

[0007] In conjunction with the first aspect mentioned above, the cleaning process uses the 3σ criterion to remove outliers, delete duplicate data, and use linear interpolation to fill in missing data.

[0008] In conjunction with the first aspect mentioned above, the load control model is a two-way mapping model, including a forward mode and a reverse mode; The forward mode takes virtual operation commands as input and outputs energy consumption parameters and production capacity parameters under the corresponding operating conditions; the reverse mode takes target energy consumption parameters as input and outputs target operation commands in reverse.

[0009] In conjunction with the first aspect above, the process of generating the actual predicted trajectory and multiple virtual predicted trajectories in parallel within the digital twin space includes: Starting from the current moment, a sliding time window mechanism is used to set the prediction duration; Based on the current operation command, the positive mode of the load control model is input to generate the actual predicted trajectory; Multiple virtual operation commands are generated and input into the positive mode of the load control model to generate corresponding virtual prediction trajectories.

[0010] In conjunction with the first aspect above, the number of virtual operation instruction generation groups is dynamically adjusted adaptively based on the real-time stability of the production line. When the variance of the equipment operating data in multiple consecutive sampling periods is lower than the first preset threshold, the production line is determined to be in a stable operating condition, and candidate vectors are generated using the first preset number of groups. When the variance of the equipment operating data in multiple consecutive sampling periods is higher than the second preset threshold, the production line is determined to be in a fluctuating condition, and a candidate vector is generated using a second preset number of values ​​that is greater than the first preset number of groups.

[0011] In conjunction with the first aspect mentioned above, the load control model is built using a reversible neural network architecture, which includes, in sequence, an input encoding layer, a reversible network layer, a regularization constraint layer, and an inverse decoding layer. The input encoding layer is used to perform high-dimensional mapping and standardized encoding of the original features of the input equipment operation data, operation instructions, capacity parameters, and energy consumption parameters. Its operational expression is as follows: in, The original input feature vector of the model. The input is the coding layer weight matrix. For bias vectors, These are the primary high-dimensional features after encoding; The reversible network layer adopts a two-branch reversible coupling structure, which splits the primary high-dimensional features into first-branch features. Features of the second branch The bidirectional invertible transformation of features is achieved through invertible mapping. The forward operation expression is: in, It is a nonlinear characteristic transformation function. These are the deep fusion features after reversible transformation; The regularization constraint layer carries regularization and monotonicity constraint operators, and the constraint expression is as follows: in, The regularization coefficient is... For the entire network weight, For physical constraint weights, As a constraint term representing the monotonic correlation between energy consumption and production capacity, output the standardized deep features after constraint. ; The inverse decoding layer is a reversible mapping reverse parsing structure. It receives the constrained depth features and decodes the target parameters and operation instructions according to the inverse solution logic. The inverse decoding expression is: in, This is the weight matrix of the inverse decoding layer. To decode the bias vector, The energy consumption parameters, production capacity parameters, or target operation commands derived from the model output.

[0012] In conjunction with the first aspect above, the training dataset of the load control model includes a historical operation dataset and a simulated operation dataset; wherein, the historical operation dataset is obtained from an industrial cloud database, and the simulated operation dataset is constructed by collecting manual adjustment operation instructions from operators when responding to different capacity demands, and then associating and grouping the manual adjustment operation instructions with the corresponding capacity demands; the manual adjustment operation instructions are the same as the operation instructions defined in the input layer of the load control model in terms of data format and dimensions.

[0013] In conjunction with the first aspect above, the method for obtaining the simulated operation dataset includes: Simulate different capacity demand scenarios of the current cement production line and simultaneously collect manual adjustment operation instructions from operators under each capacity demand scenario. The collected manual adjustment operation commands are processed by outlier removal, smoothing filtering and normalization to obtain operation commands that can be directly input into the encoding layer of the load control model; By binding different production capacity demands with corresponding standardized operating instructions one by one, training sample pairs are formed, thus completing the construction of the simulated operation dataset.

[0014] Secondly, the present invention provides a cement production line workload control system based on a large model, the system comprising: The acquisition module is used to acquire real-time equipment operation data of the cement production line. The generation module is used to generate, in parallel, actual predicted trajectories and multiple virtual predicted trajectories in the digital twin space based on a preset load control model when the device operation data changes. The actual predicted trajectory is the predicted trajectory that maintains the current operation command, and the virtual predicted trajectory is the predicted trajectory after adjusting at least one operation command. The module is used to construct a coordinate system with energy consumption parameters as the X-axis and production capacity parameters as the Y-axis, and to extract the energy consumption parameters and production capacity parameters of each predicted trajectory endpoint to form a coordinate set; The fitting module is used to compare all coordinate data in the coordinate set pairwise, remove the inferior coordinates in the set that simultaneously satisfy the condition that the energy consumption parameter is greater than the other coordinates and the production capacity parameter is less than the other coordinates, and connect the filtered coordinates in sequence to form a fitting curve. The output module is used to determine whether the coordinates corresponding to the actual predicted trajectory are located on the fitted curve. If yes, the current operation command remains unchanged; otherwise, the real-time capacity demand is obtained, and the target energy consumption parameters are solved in reverse by combining the fitted curve. The target energy consumption parameters are then input into a preset load control model to obtain the target operation command.

[0015] Thirdly, the present invention provides a computer device including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor performs the above-described method for controlling the workload of a cement production line based on a large model.

[0016] Fourthly, the present invention provides a computer-readable storage medium storing a plurality of computer programs that can be loaded by a processor to execute any of the large-model-based cement production line workload control methods provided in the embodiments of the present invention. The computer program can execute the following steps of the large-model-based cement production line workload control method: Real-time acquisition of equipment operation data from cement production lines; When the device's operating data changes, an actual predicted trajectory and multiple virtual predicted trajectories are generated in parallel in the digital twin space based on a preset load control model. The actual predicted trajectory is the predicted trajectory that maintains the current operation command, and the virtual predicted trajectory is the predicted trajectory after adjusting at least one operation command. A coordinate system is constructed with energy consumption parameters as the X-axis and production capacity parameters as the Y-axis. The energy consumption parameters and production capacity parameters at the endpoints of each predicted trajectory are extracted to form a coordinate set. All coordinate data in the coordinate set are compared pairwise, and the inferior coordinates that simultaneously satisfy the condition that the energy consumption parameter is greater than the other coordinates and the production capacity parameter is less than the other coordinates are removed. The filtered coordinates are then connected to form a fitting curve. Determine whether the coordinates corresponding to the actual predicted trajectory are located on the fitted curve. If yes, keep the current operation command unchanged. If no, obtain the real-time capacity demand, combine the fitted curve to solve the target energy consumption parameters in reverse, and input the target energy consumption parameters into the preset load control model to obtain the target operation command.

[0017] In summary, the cement production line workload control method based on a large model provided by this invention, through the use of simulation operation and a load control model, can objectively and quantitatively reveal the trade-off boundary between energy consumption and production capacity under the current operating conditions, achieving a balance between energy consumption and production capacity, while significantly reducing trial-and-error costs. It adaptively adjusts the number of virtual operation command generation groups according to the stability of the production line's real-time operating conditions. Under stable operating conditions, fewer groups are used to reduce computational overhead, while under fluctuating operating conditions, more groups are used to fully explore the optimal strategy. This dynamic adjustment mechanism ensures optimization effectiveness while avoiding waste of computational resources. A reversible neural network architecture is used to build the load control model, which is similar to... It supports both forward and reverse modes. The target energy consumption parameters obtained by inversely solving the fitted curve can be directly input into the reverse mode to obtain executable operation instructions, forming a complete closed loop of "capacity demand - optimal energy consumption - operation instructions" to achieve subsequent precise control. The training dataset not only includes historical operation data obtained from the industrial cloud, but also constructs a simulated operation dataset by simulating different capacity demand scenarios and collecting manual adjustment instructions from operators. This process quantifies the implicit experience knowledge accumulated by senior operators over many years into training sample pairs that can be associated and bound with capacity demand, enabling the load control model to have a high level of control in the early stage of production line commissioning and shortening the on-site debugging cycle. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0019] To gain a more complete understanding of the present invention and its beneficial effects, the following description will be made in conjunction with the accompanying drawings, wherein the same reference numerals denote the same parts in the following description.

[0020] Figure 1 This is a flowchart of a cement production line workload control method based on a large model in an exemplary embodiment of the present invention.

[0021] Figure 2 This is a schematic diagram of the working load control system for a cement production line based on a large model, according to an exemplary embodiment of the present invention.

[0022] Figure 3 This is a schematic diagram of the structure of a computer device in an exemplary embodiment of the present invention. Detailed Implementation

[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the protection scope of the present invention.

[0024] Figure 1 This is an exemplary embodiment of the cement production line workload control method based on a large model provided by the present invention. The process steps of this large model-based cement production line workload control method are merely one possible implementation of the present invention. Figure 1 As shown, the cement production line workload control method and system based on a large model provided in this embodiment of the invention includes the following steps: S10 acquires real-time equipment operation data from the cement production line.

[0025] S20, when the device operating data changes, an actual predicted trajectory and multiple virtual predicted trajectories are generated in parallel in the digital twin space based on a preset load control model. The actual predicted trajectory is the predicted trajectory that maintains the current operation command, and the virtual predicted trajectory is the predicted trajectory after adjusting at least one operation command.

[0026] In this embodiment, the digital twin space is a complete mapping of the physical cement production line in the virtual digital world. It has been applied in multiple fields, including the cement industry, and is a real-time, interactive, and predictable dynamic simulation environment. By introducing the digital twin space, this embodiment can greatly reduce the cost of trial and error. The predicted trajectory refers to a continuous, timestamped path of state change obtained in the digital twin space, starting from the current moment and extending into the future.

[0027] S30: Construct a coordinate system with energy consumption parameters as the X-axis and production capacity parameters as the Y-axis, and extract the energy consumption parameters and production capacity parameters of each predicted trajectory endpoint to form a coordinate set.

[0028] In this embodiment, the timestamps of the endpoints of each predicted trajectory are the same, meaning they all experienced the same prediction duration.

[0029] S40, perform pairwise comparisons on all coordinate data in the coordinate set, eliminate the inferior coordinates in the set that simultaneously satisfy the condition that the energy consumption parameter is greater than the other coordinates and the production capacity parameter is less than the other coordinates, and connect the selected coordinates in sequence to form a fitting curve.

[0030] S50, determine whether the coordinates corresponding to the actual predicted trajectory are located on the fitted curve. If yes, keep the current operation command unchanged. If no, obtain the real-time capacity demand, combine the fitted curve to solve the target energy consumption parameters in reverse, and input the target energy consumption parameters into the preset load control model to obtain the target operation command.

[0031] In this embodiment, if the coordinates corresponding to the actual predicted trajectory are located on the fitted curve, it indicates that the change command does not have a significant advantage over the maintain command. This can keep the current operation command unchanged in some cases, thereby preventing unnecessary adjustment commands from being issued frequently due to minor disturbances, avoiding frequent fluctuations in the production line, and improving operational stability.

[0032] In summary, the cement production line workload control method based on a large model provided in this embodiment of the invention, by using simulation operation and load control model, can objectively and quantitatively reveal the trade-off boundary between energy consumption and production capacity under the current operating conditions, achieve a balance between energy consumption and production capacity, and greatly reduce trial and error costs.

[0033] In one embodiment of the present invention, the real-time acquisition of equipment operation data of the cement production line includes: S101, real-time acquisition of raw multi-source data from the cement production line, including equipment operating parameters, video images, and audio signals. In this embodiment, a multi-source data acquisition method is adopted, which can more accurately characterize the operating status of the equipment. Specifically, video images of thermal equipment (such as rotary kilns) can be processed to extract the internal flame temperature characteristics, while sound signals of mechanical equipment (such as mills) can extract load characteristics.

[0034] S102, the original multi-source data is cleaned to obtain preprocessed multi-source data.

[0035] In this embodiment, cleaning the original multi-source data can reduce noise and improve data accuracy.

[0036] S103, Time-stamp alignment is performed on the preprocessed multi-source data to obtain the device operation data.

[0037] Furthermore, in the above embodiments, the cleaning process employs the 3σ criterion to remove outliers, delete duplicate data, and use linear interpolation to fill in missing data. Specifically, regarding the 3σ criterion, for any device's operating parameter sequence x1, x2, ..., x... n Calculate its mean and standard deviation The expression is: When a data point satisfies If so, it is determined to be an outlier and is removed; The deletion of duplicate data is based on timestamps; if they are exactly the same, they are considered duplicates. Regarding linear interpolation, for a certain time t i Missing value x i Using its effective data points x before and after i-1 and x i+1 Linear interpolation is performed, and the specific expression is as follows: .

[0038] In one embodiment of the present invention, the load control model is a bidirectional mapping model, including a forward mode and a reverse mode; The forward mode takes virtual operation commands as input and outputs energy consumption parameters and production capacity parameters under the corresponding operating conditions; the reverse mode takes target energy consumption parameters as input and outputs target operation commands in reverse.

[0039] In one embodiment of the present invention, the process of generating actual predicted trajectories and multiple virtual predicted trajectories in parallel within the digital twin space includes: S201 uses the current moment as the starting point and a sliding time window mechanism to set the prediction duration.

[0040] In this embodiment, the sliding time window mechanism refers to taking the current moment as the starting point and extending a fixed time into the future as the prediction interval, which dynamically slides forward as time progresses.

[0041] S202, based on the current operation command, input the positive mode of the load control model to generate the actual predicted trajectory.

[0042] S203, generate multiple virtual operation commands, input them into the positive mode of the load control model respectively, and generate multiple corresponding virtual prediction trajectories.

[0043] Furthermore, in the above embodiments, the number of virtual operation instruction generation groups is dynamically adjusted adaptively according to the real-time stability of the production line. When the variance of the equipment operating data in multiple consecutive sampling periods is lower than the first preset threshold, the production line is determined to be in a stable operating condition, and candidate vectors are generated using the first preset number of groups. When the variance of the equipment operating data in multiple consecutive sampling periods is higher than the second preset threshold, the production line is determined to be in a fluctuating condition, and a candidate vector is generated using a second preset number of values ​​that is greater than the first preset number of groups.

[0044] Specifically, assuming the length of the sliding window is L, the variance of any key parameter x is calculated using the following formula: in, The arithmetic mean of the parameters within the window is expressed as: By incorporating multiple key parameters, the overall variance is defined as: in, The total number of key parameters involved in the evaluation. The weighting coefficients of each parameter are as follows: , The normalized baseline variance of the j-th parameter is taken as the statistical value under historical stable operating conditions. The comprehensive variance is calculated using the above formula and then compared with the first preset threshold and the second preset threshold. The first preset threshold and the second preset threshold can be set according to the actual working conditions.

[0045] In summary, the embodiments of the present invention adaptively adjust the number of virtual operation instruction generation groups according to the stability of the real-time operating conditions of the production line. Under stable operating conditions, fewer groups are used to reduce computational overhead, while under fluctuating operating conditions, more groups are used to fully explore the optimal strategy. This dynamic adjustment mechanism ensures the optimization effect while avoiding the waste of computing resources.

[0046] In one embodiment of the present invention, the load control model is built using a reversible neural network architecture, which includes, in sequence, an input encoding layer, a reversible network layer, a regularization constraint layer, and a reverse decoding layer.

[0047] Traditional deep neural networks typically map in a unidirectional and irreversible manner. In contrast, reversible neural networks, through a special network structure design, make forward and reverse mappings inverse operations, while also achieving high computational accuracy and lossless information. This embodiment utilizes this characteristic to meet the load control requirements.

[0048] Specifically, the input encoding layer is used to perform high-dimensional mapping and standardized encoding of the original features of the input equipment operation data, operation instructions, production capacity parameters, and energy consumption parameters. Its operational expression is as follows: in, The original input feature vector of the model. The input is the coding layer weight matrix. For bias vectors, These are the primary high-dimensional features after encoding; The reversible network layer adopts a two-branch reversible coupling structure, which splits the primary high-dimensional features into first-branch features. Features of the second branch The bidirectional invertible transformation of features is achieved through invertible mapping. The forward operation expression is: in, It is a nonlinear characteristic transformation function. These are the deep fusion features after reversible transformation; The regularization constraint layer carries regularization and monotonicity constraint operators to apply physical constraints to the deep fusion features output by the invertible network layer. This ensures that the energy consumption-energy production relationship output by the model conforms to common sense in engineering and avoids predictions that violate physical laws. The constraint expression is as follows: in, The regularization coefficient is... For the entire network weight, For physical constraint weights, As a constraint term representing the monotonic correlation between energy consumption and production capacity, output the standardized deep features after constraint. ; The inverse decoding layer is a reversible mapping reverse parsing structure. It receives the constrained depth features and decodes the target parameters and operation instructions according to the inverse solution logic. The inverse decoding expression is: in, This is the weight matrix of the inverse decoding layer. To decode the bias vector, The energy consumption parameters, production capacity parameters, or target operation commands derived from the model output.

[0049] In summary, the load control model in this embodiment adopts a reversible neural network architecture to build the load control model. This model supports both forward and reverse modes. Combined with the target energy consumption parameters obtained by inversely solving the fitted curve, the reverse mode can be directly input to obtain executable operation instructions, forming a complete closed loop of "capacity demand - optimal energy consumption - operation instructions" to achieve subsequent precise regulation.

[0050] In one embodiment of the present invention, the training dataset of the load control model includes a historical operation dataset and a simulated operation dataset; wherein, the historical operation dataset is obtained from an industrial cloud database, and the simulated operation dataset is constructed by collecting manual adjustment operation instructions from operators when responding to different capacity demands, and associating the manual adjustment operation instructions with the corresponding capacity demands; the manual adjustment operation instructions are the same as the operation instructions defined in the input layer of the load control model in terms of data format and dimensions.

[0051] This embodiment defines the composition of the training dataset for the load control model, which includes two main types of data sources: historical operation datasets and simulated operation datasets. The historical operation datasets are obtained from industrial cloud databases. Although historical operation datasets can reflect real production conditions, they may have incomplete coverage, such as some extreme working conditions. The purpose of simulated operation datasets is to make up for the above deficiencies. By actively constructing training samples, the learning effect of the model can be enhanced. At the same time, the implicit experience knowledge of operators can also be integrated into the model learning.

[0052] Furthermore, in the above embodiments, the method for obtaining the simulated operation dataset includes: S100 simulates different capacity demand scenarios of the current cement production line and simultaneously collects manual adjustment operation instructions from operators under each capacity demand scenario.

[0053] In this embodiment, for the typical capacity range of the mud production line (e.g., 60%~110% of the designed capacity), discrete capacity demand values ​​are generated in steps of 5%~10%. For each capacity demand value, different operating disturbance conditions (such as different raw material grindability and different ambient temperatures) are further superimposed to form diverse simulation scenarios. The above simulation scenarios are presented to experienced operators, who are required to adjust the operation instructions to make the production line reach the target capacity while minimizing energy consumption. Then, the manual adjustment operation instructions of the operators in each scenario are collected simultaneously.

[0054] S200 performs outlier removal, smoothing filtering, and normalization on the collected manual adjustment operation commands to obtain operation commands that can be directly input into the encoding layer of the load control model.

[0055] S300 binds different production capacity demands with corresponding standardized operating instructions one by one to form training sample pairs, thus completing the construction of the simulated operation dataset.

[0056] In summary, the training dataset of the load control model in this embodiment of the invention not only includes historical operation data obtained from the industrial cloud, but also constructs a simulated operation dataset by simulating different capacity demand scenarios and collecting manual adjustment instructions from operators. This process quantifies the implicit experience knowledge accumulated by senior operators over many years into training sample pairs that can be associated with and bound to capacity demand, enabling the load control model to have a high level of control in the early stage of production line commissioning and shortening the on-site debugging cycle.

[0057] Figure 2 This is an exemplary embodiment of the cement production line workload control system based on a large model provided by the present invention. This large-model-based cement production line workload control system is merely one possible implementation of the present invention. Figure 2 As shown, the cement production line workload control system based on a large model provided in this embodiment of the invention includes: The acquisition module is used to acquire real-time equipment operation data of the cement production line. The generation module is used to generate, in parallel, actual predicted trajectories and multiple virtual predicted trajectories in the digital twin space based on a preset load control model when the device operation data changes. The actual predicted trajectory is the predicted trajectory that maintains the current operation command, and the virtual predicted trajectory is the predicted trajectory after adjusting at least one operation command. The module is used to construct a coordinate system with energy consumption parameters as the X-axis and production capacity parameters as the Y-axis, and to extract the energy consumption parameters and production capacity parameters of each predicted trajectory endpoint to form a coordinate set; The fitting module is used to compare all coordinate data in the coordinate set pairwise, remove the inferior coordinates in the set that simultaneously satisfy the condition that the energy consumption parameter is greater than the other coordinates and the production capacity parameter is less than the other coordinates, and connect the filtered coordinates in sequence to form a fitting curve. The output module is used to determine whether the coordinates corresponding to the actual predicted trajectory are located on the fitted curve. If yes, the current operation command remains unchanged; otherwise, the real-time capacity demand is obtained, and the target energy consumption parameters are solved in reverse by combining the fitted curve. The target energy consumption parameters are then input into a preset load control model to obtain the target operation command.

[0058] Figure 3 This is a schematic diagram of the structure of the computer device provided in an embodiment of the present invention, such as... Figure 3As shown, the computer device includes a memory and a processor. The memory stores a computer program, which, when executed by the processor, causes the processor to perform the cement production line workload control method and system based on a large model as described in the above embodiments.

[0059] Furthermore, those skilled in the art will understand that all or part of the steps in the various large-model-based cement production line workload control methods and systems described above can be accomplished by instructions, or by controlling related hardware through instructions. These instructions can be stored in a computer-readable storage medium and loaded and executed by a processor.

[0060] To this end, embodiments of the present invention provide a computer-readable storage medium storing a plurality of computer programs. These computer programs can be loaded by a processor to execute any of the large-model-based cement production line workload control methods provided in the embodiments of the present invention. The computer program can execute the following steps of the large-model-based cement production line workload control method: S10 acquires real-time equipment operation data from the cement production line.

[0061] S20, when the device operating data changes, an actual predicted trajectory and multiple virtual predicted trajectories are generated in parallel in the digital twin space based on a preset load control model. The actual predicted trajectory is the predicted trajectory that maintains the current operation command, and the virtual predicted trajectory is the predicted trajectory after adjusting at least one operation command.

[0062] In this embodiment, the digital twin space is a complete mapping of the physical cement production line in the virtual digital world. It has been applied in multiple fields, including the cement industry, and is a real-time, interactive, and predictable dynamic simulation environment. By introducing the digital twin space, this embodiment can greatly reduce the cost of trial and error. The predicted trajectory refers to a continuous, timestamped path of state change obtained in the digital twin space, starting from the current moment and extending into the future.

[0063] S30: Construct a coordinate system with energy consumption parameters as the X-axis and production capacity parameters as the Y-axis, and extract the energy consumption parameters and production capacity parameters of each predicted trajectory endpoint to form a coordinate set.

[0064] S40, perform pairwise comparisons on all coordinate data in the coordinate set, eliminate the inferior coordinates in the set that simultaneously satisfy the condition that the energy consumption parameter is greater than the other coordinates and the production capacity parameter is less than the other coordinates, and connect the selected coordinates in sequence to form a fitting curve.

[0065] S50, determine whether the coordinates corresponding to the actual predicted trajectory are located on the fitted curve. If yes, keep the current operation command unchanged. If no, obtain the real-time capacity demand, combine the fitted curve to solve the target energy consumption parameters in reverse, and input the target energy consumption parameters into the preset load control model to obtain the target operation command. The computer program instructions described in this application embodiment may be stored in a computer-readable storage medium capable of directing a computer or other programmable data processing device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, the instruction means being implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0066] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0067] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0068] Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, like read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0069] Computer-readable media include both permanent and non-permanent, removable and non-removable media, which can store information using any model-based method, system, or technology for controlling the workload of a cement production line. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient media, such as modulated communication signals and carrier waves.

[0070] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0071] The embodiments, implementation methods, and related technical features of this invention can be combined and substituted for each other without conflict.

[0072] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any way. Any simple modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for controlling the workload of a cement production line based on a large model, characterized in that, include: Real-time acquisition of equipment operation data from cement production lines; When the device's operating data changes, an actual predicted trajectory and multiple virtual predicted trajectories are generated in parallel in the digital twin space based on a preset load control model. The actual predicted trajectory is the predicted trajectory that maintains the current operation command, and the virtual predicted trajectory is the predicted trajectory after adjusting at least one operation command. A coordinate system is constructed with energy consumption parameters as the X-axis and production capacity parameters as the Y-axis. The energy consumption parameters and production capacity parameters at the endpoints of each predicted trajectory are extracted to form a coordinate set. All coordinate data in the coordinate set are compared pairwise, and the inferior coordinates that simultaneously satisfy the condition that the energy consumption parameter is greater than the other coordinates and the production capacity parameter is less than the other coordinates are removed. The filtered coordinates are then connected to form a fitting curve. Determine whether the coordinates corresponding to the actual predicted trajectory are located on the fitted curve. If yes, keep the current operation command unchanged. If no, obtain the real-time capacity demand, combine the fitted curve to solve the target energy consumption parameters in reverse, and input the target energy consumption parameters into the preset load control model to obtain the target operation command.

2. The method for controlling the workload of a cement production line based on a large model according to claim 1, characterized in that, The real-time acquisition of equipment operation data for the cement production line includes: Real-time acquisition of raw multi-source data from the cement production line, including equipment operating parameters, video images, and audio signals; The original multi-source data is cleaned to obtain preprocessed multi-source data; The preprocessed multi-source data is time-stamped to obtain the device operation data.

3. The method for controlling the workload of a cement production line based on a large model according to claim 2, characterized in that, The cleaning process uses the 3σ criterion to remove outliers, delete duplicate data, and use linear interpolation to fill in missing data.

4. The method for controlling the workload of a cement production line based on a large model according to claim 1, characterized in that, The load control model is a two-way mapping model, including forward mode and reverse mode; The forward mode takes virtual operation commands as input and outputs energy consumption parameters and production capacity parameters under the corresponding operating conditions; the reverse mode takes target energy consumption parameters as input and outputs target operation commands in reverse.

5. The method for controlling the workload of a cement production line based on a large model according to claim 4, characterized in that, The process of generating the actual predicted trajectory and multiple virtual predicted trajectories in parallel within the digital twin space includes: Starting from the current moment, a sliding time window mechanism is used to set the prediction duration; Based on the current operation command, the positive mode of the load control model is input to generate the actual predicted trajectory; Multiple virtual operation commands are generated and input into the positive mode of the load control model to generate corresponding virtual prediction trajectories.

6. The method for controlling the workload of a cement production line based on a large model according to claim 5, characterized in that, The number of virtual operation instruction generation groups is dynamically adjusted adaptively based on the real-time stability of the production line. When the variance of the equipment operating data in multiple consecutive sampling periods is lower than the first preset threshold, the production line is determined to be in a stable operating condition, and candidate vectors are generated using the first preset number of groups. When the variance of the equipment operating data in multiple consecutive sampling periods is higher than the second preset threshold, the production line is determined to be in a fluctuating condition, and a candidate vector is generated using a second preset number of values ​​that is greater than the first preset number of groups.

7. The method for controlling the workload of a cement production line based on a large model according to claim 6, characterized in that, The load control model is built using a reversible neural network architecture, which includes an input encoding layer, a reversible network layer, a regularization constraint layer, and a reverse decoding layer. The input encoding layer is used to perform high-dimensional mapping and standardized encoding of the original features of the input equipment operation data, operation instructions, capacity parameters, and energy consumption parameters. Its operational expression is as follows: in, The original input feature vector of the model. The input is the coding layer weight matrix. For bias vectors, These are the primary high-dimensional features after encoding; The reversible network layer adopts a two-branch reversible coupling structure, which splits the primary high-dimensional features into first-branch features. Features of the second branch The bidirectional invertible transformation of features is achieved through invertible mapping. The forward operation expression is: in, It is a nonlinear characteristic transformation function. These are the deep fusion features after reversible transformation; The regularization constraint layer carries regularization and monotonicity constraint operators, and the constraint expression is as follows: in, The regularization coefficient is... For the entire network weight, For physical constraint weights, As a constraint term representing the monotonic correlation between energy consumption and production capacity, output the standardized deep features after constraint. ; The inverse decoding layer is a reversible mapping reverse parsing structure. It receives the constrained depth features and decodes the target parameters and operation instructions according to the inverse solution logic. The inverse decoding expression is: in, This is the weight matrix of the inverse decoding layer. To decode the bias vector, The energy consumption parameters, production capacity parameters, or target operation commands derived from the model output.

8. The method for controlling the workload of a cement production line based on a large model according to claim 7, characterized in that, The training dataset of the load control model includes a historical operation dataset and a simulated operation dataset. The historical operation dataset is obtained from an industrial cloud database, and the simulated operation dataset is constructed by collecting manual adjustment operation instructions from operators when responding to different capacity demands, and then associating and grouping the manual adjustment operation instructions with the corresponding capacity demands. The manual adjustment operation instructions are the same as the operation instructions defined in the input layer of the load control model in terms of data format and dimensions.

9. The method for controlling the workload of a cement production line based on a large model according to claim 8, characterized in that, The method for obtaining the simulation operation dataset includes: Simulate different capacity demand scenarios of the current cement production line and simultaneously collect manual adjustment operation instructions from operators under each capacity demand scenario. The collected manual adjustment operation commands are processed by outlier removal, smoothing filtering and normalization to obtain operation commands that can be directly input into the encoding layer of the load control model; By binding different production capacity demands with corresponding standardized operating instructions one by one, training sample pairs are formed, thus completing the construction of the simulated operation dataset.

10. A working load control system for a cement production line based on a large model, characterized in that, include: The acquisition module is used to acquire real-time equipment operation data of the cement production line. The generation module is used to generate, in parallel, actual predicted trajectories and multiple virtual predicted trajectories in the digital twin space based on a preset load control model when the device operation data changes. The actual predicted trajectory is the predicted trajectory that maintains the current operation command, and the virtual predicted trajectory is the predicted trajectory after adjusting at least one operation command. The module is used to construct a coordinate system with energy consumption parameters as the X-axis and production capacity parameters as the Y-axis, and to extract the energy consumption parameters and production capacity parameters of each predicted trajectory endpoint to form a coordinate set; The fitting module is used to compare all coordinate data in the coordinate set pairwise, remove the inferior coordinates in the set that simultaneously satisfy the condition that the energy consumption parameter is greater than the other coordinates and the production capacity parameter is less than the other coordinates, and connect the filtered coordinates in sequence to form a fitting curve. The output module is used to determine whether the coordinates corresponding to the actual predicted trajectory are located on the fitted curve. If yes, the current operation command remains unchanged; otherwise, the real-time capacity demand is obtained, and the target energy consumption parameters are solved in reverse by combining the fitted curve. The target energy consumption parameters are then input into a preset load control model to obtain the target operation command.

Images